search

angelolab / Nimbus-Inference

MIT License
4 stars 3 forks source link

NaNs in predictions #14

Open ngreenwald opened 2 weeks ago

ngreenwald commented 2 weeks ago

Report

When running the model on square images, @peterjchou was getting an error where all of the predictions across all the cells were NaNs.

Peter, could you attach the image you used that generated the issue so we can investigate? Or if it's quite large, maybe a small crop of it, if you see the same issue there?

Thanks!

Version information

No response

peterjchou commented 2 weeks ago

Thanks for starting this thread, Noah! I'm attaching the aSMA image and whole cell segmentation from fov1 that produced the attached cell_table. These are 8bit CODEX images. nimbus.zip

cell_table_screenshot
JLrumberger commented 1 week ago

Hi @peterjchou ,

I ran inference on the FoV you provided and don't get NaNs. My guess is that something is wrong with the normalization. Could you send me the normalization_dict.json in the nimbus_output folder?

Best, Lorenz

peterjchou commented 1 week ago

Thanks for looking into this Lorenz, I appreciate the help!

When you ran inference on the FoV did you also get visible predictions? I just ran it using only this FoV and didn't get NaNs this time, but the predictions were artifacts.

normalization_dict.json

nimbus_viewer_output
JLrumberger commented 1 week ago

Your normalization_dict.json looks alright, but the prediction does not look correct. Here's my output: image

I pushed some fixes to the main branch. Could you pull the main branch and re-run the notebook? If you have problems with this, I'd be happy to jump on a call and help you.